Automatic Pulmonary Nodule Detection in CT Scans Using Convolutional Neural Networks Based on Maximum Intensity...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate pulmonary nodule detection is vital for lung cancer screening.
• Computer-aided detection (CAD) systems are underutilized in clinical practice.
• Maximum intensity projection (MIP) images can aid nodule detection in CT scans.

Purpose of the Study:

• To explore the use of MIP images to improve automatic lung nodule detection with convolutional neural networks (CNNs).
• To enhance the effectiveness of nodule detection by integrating clinical methodology into AI frameworks.

Main Methods:

• A CNN-based approach utilizing MIP images (5, 10, 15 mm slab thicknesses) and 1 mm axial CT slices.
• Augmenting 2-D CT slices with spatial information from MIP images to differentiate nodules from vessels.
• Utilizing the LIDC-IDRI dataset comprising 888 scans.

Main Results:

• Achieved 92.7% sensitivity with 1 false positive per scan and 94.2% sensitivity with 2 false positives per scan.
• Demonstrated improved detection of small pulmonary nodules (3-10 mm) using thick MIP images.
• Showcased a reduction in false positives compared to methods without MIP images.

Conclusions:

• MIP-based CNNs significantly improve sensitivity and reduce false positives for pulmonary nodule detection in CT scans.
• Integrating clinical procedures like MIP imaging enhances AI performance in medical diagnostics.
• The proposed framework shows promise for routine clinical application in lung cancer screening.